Human activity recognition (HAR) research using smartphone sensor technology has changed people's daily lives and opened the door for numerous exciting applications due to its extensive applications in addressing real-world humancentric issues. These studies, however, mostly concentrated on a woman's daily activities and behaviours when she is attacked, such as running, walking, sleeping, eating, drinking, lying, straight punching, front kicking, and so forth. The article offers a method to foresee human behaviour using a deep learning model and evaluates the efficiency of the approach using real-world data in response to this. We offer an LSTM structure for detecting the intensity of HAR in order to track activity patterns and build a neural network with binary classification skills to evaluate the situation of women. The proposed model uses the dataset to identify human activity, which records 30 volunteers' body movements while they engage in 18 physical activities. Feature engineering is used to extract additional variables from the data in addition to that. Our proposed model achieves a 90.89% accuracy. © 2023 IEEE.